How to Use Claude Code Commands More Efficiently

TL;DR

Use /btw to ask quick questions while a long task runs, and control token consumption by matching the model and reasoning effort to each job. Keep an efficient default configuration, then use /advisor or the ultrathink keyword when a difficult prompt needs stronger judgment, while /context and optional plugins help monitor or reduce context usage.

Transcript

I'm going to share with you 19 secret codes for Claude Code to 10X your productivity in the next 45 minutes. I teach AI to millions of people every month. I hit 30 million views last month and I've tested all the Claude commands to find the ones most people have no idea exist. And these work whether you're new to Claude or you've been using it for ... Read More

Key Insights

• /btw is a command for asking a quick question while Claude Code is still completing a long-running task. It preserves the current conversation and context, so the user does not need to create another session or open another tab for the smaller request.
• /model is the control for selecting among available models according to the task. The presenter recommends reserving the most advanced option for complex, technical, or sensitive work and using lighter models for straightforward tasks such as organizing local files or renaming them.
• /effort changes how much reasoning Claude applies through low, medium, high, and max settings. Lower effort reduces repeated thinking iterations, while higher effort is intended for work that needs additional reasoning, accuracy, or reflection and therefore may consume more tokens.
• /advisor is designed to provide stronger judgment when Claude encounters a difficult decision, unclear debugging problem, or stalled line of investigation. It can be configured with an advanced model and maximum reasoning while the ordinary default remains a more efficient model and effort level.
• The ultrathink keyword requests the best available model and maximum reasoning effort for a particular prompt. It can appear at the beginning, middle, or end of that prompt, allowing one difficult task to receive extra computational attention without manually changing the default model and effort settings.
• The Caveman plugin is an optional open-source addition rather than a built-in Claude Code skill. According to the presenter, running /caveman removes output that is considered unnecessary for continued performance, reducing output-token consumption while retaining the context needed to proceed.
• Context Mode is an optional open-source plugin intended to reduce token consumption from MCP tool calls. The presenter says it works in the background after installation and permission are completed, so users do not need to remember or enter a separate command during later tool use.
• /context visualizes current context consumption as a colored grid and estimates usage by category, including prompts and tools. The presenter describes it as a terminal feature and expresses uncertainty about Visual Studio Code support, so availability may depend on the interface being used.

Questions & Answers

Q: How can I use Claude Code while a long task is running?

Type /btw while the long-running prompt is still being processed, then enter the quick question you want answered. The command lets Claude address that smaller question without requiring another tab or a separate session. The original conversation and its context remain available while the larger task continues, making /btw useful when research or another extended operation is already underway.

Q: How do I choose the right Claude Code model for a task?

Use /model to select a model that matches the difficulty and sensitivity of the work. The presenter recommends advanced models for complex, technical, or highly sensitive tasks where mistakes or hallucinations are especially undesirable. Simpler work, including organizing local files, renaming items, or drafting an ordinary social post, may be assigned to a lighter model to conserve tokens.

Q: What does the /effort command do in Claude Code?

The /effort command changes the amount of reasoning applied to a task, with low, medium, high, and max levels described in the transcript. A lower setting limits the number of thinking and reflection iterations, which can reduce token use. Higher settings are more appropriate when a task needs additional reasoning, accuracy, or careful consideration beyond the user's efficient default configuration.

Q: When should I use /advisor in Claude Code?

Use /advisor when a task needs stronger judgment, such as a complex decision, an unclear debugging problem, or a situation where Claude keeps circling without progress. The feature can escalate the difficult issue to a configured advanced model with maximum reasoning. This allows the regular session to retain a more efficient default while difficult moments receive additional capability and tokens.

Q: What is the difference between /advisor and ultrathink?

/advisor is presented as an escalation mechanism that can involve a configured advanced model when Claude determines stronger judgment is needed. Ultrathink is a keyword the user adds directly to a particular prompt to request the best available model and maximum reasoning. Both support an efficient default, but ultrathink represents an explicit per-prompt request rather than automatic escalation when Claude struggles.

Q: How does the Caveman plugin reduce Claude Code token usage?

The Caveman plugin is described as an open-source plugin that must be installed because it is not a built-in Claude Code skill. After installation, the user runs /caveman to remove parts of the output that the plugin considers unnecessary for continued performance. The intended result is substantially lower output-token consumption while preserving enough context for Claude to keep working effectively.

Q: How does Context Mode help with MCP tool calls?

Context Mode is described as an open-source plugin that reduces token consumption associated with MCP tool calls. The user can ask Claude to install it and provide the necessary permission. According to the presenter, no recurring command is required afterward because the plugin operates in the background, making it a complementary option to /caveman for users on a lower-tier plan.

Q: How can I check current context usage in Claude Code?

Run /context to display current context consumption as a colored grid and view estimated usage divided into categories such as prompts and tools. This helps reveal how the available context is being used rather than treating it as an invisible total. The presenter believes the command is limited to the terminal at the time discussed and may not work in Visual Studio Code.

Summary & Key Takeaways

• Claude Code can remain useful during a long-running task through /btw, which answers a separate quick question within the same conversation and context. The command avoids opening another tab or starting another session merely to handle a short request while research, competitor analysis, or another extended operation continues.

• Token management begins with choosing an appropriate model and reasoning effort. The presenter recommends lighter options for simple work, such as organizing or renaming files, and stronger models for complex, technical, or sensitive tasks. A lower-cost default can then be temporarily escalated when a particular problem requires deeper reasoning.

• Several features support selective escalation and context control. /advisor can involve a configured advanced model when stronger judgment is needed, while ultrathink requests the best available model and maximum reasoning for one prompt. The Caveman plugin reduces unnecessary output, Context Mode targets MCP tool-call consumption, and /context visualizes current usage.